Association for Heterodox Economics 9th Annual Conference 2007

University of West England, Bristol, UK

13 – 15 July 2007

Grounded theory and the study of retirement savings: A case studyof broadening methods applied to economic research projects

Therese JeffersonWomen in Social & Economic Research (WiSER) Unit

Curtin University of TechnologyGraduate School of Business

PO Box U1987,Perth, Western Australia, 6845

Email: T.Jefferson@curtin.edu.au

AbstractRecent discussions about economics and ontology, have generated debate aboutmethods that are appropriate for economic research that endeavours to be consistentwith critical realism. Grounded theory is one research method that has been identifiedas potentially relevant but so far it has been used by only a few researchers within thefield of economics.

This paper provides a case study of using grounded theory as a research method in aneconomic research project. The paper consists of four main sections. The first sectionprovides an outline of the key features of grounded theory and the reasons that itmight be identified as largely consistent with a critical realist approach to economicresearch. Grounded theory’s approach to theory construction based on data isidentified as particularly relevant. The second section consists of a description of theapplication of grounded theory to a research project examining women’s patterns ofsaving for retirement in Western Australia. This includes a discussion of key aspectsof the project including: definition of the research question; identifying and collectingdata; analysis; and presentation of findings. The third and fourth sections provide anassessment of some advantages and disadvantages of utilising grounded theory in thecontext of a broader economic research project.

Two key conclusions are drawn. Firstly, there are considerable challenges to be metwhen employing grounded theory in an economic research project. These challengesarise from both the history and nature of grounded theory itself and from the relativelack of understanding among many economists and policy makers of the goals andprocesses involved in grounded theory research projects. Secondly and perhaps moreimportantly, grounded theory provides a useful way of generating potentially usefulinsights that are not generally obtained from the more standard array of economicresearch methods.

1

Grounded theory and the study of retirement savings: A case study ofbroadening methods applied in economic research projects

1. IntroductionRecent debates about connections between economic theory and ontologicalapproaches, particularly critical realism, have generated a range of debates. Oneimportant aspect of these debates involves the identification of methods that areappropriate for economic research that endeavours to be consistent with criticalrealism. Grounded theory is one research method that has been identified aspotentially relevant to those with an interest in critical realism but so far it has beenused by only a few researchers within the field of economics.

This paper provides a case study of using grounded theory as a research method in aneconomic research project. The paper consists of four main sections. The first sectionprovides an outline of the key features of grounded theory and the reasons that itmight be identified as largely consistent with a critical realist approach to economicresearch. Grounded theory’s approach to theory construction based on data isidentified as particularly relevant. The second section consists of a description of theapplication of grounded theory to a research project examining women’s patterns ofsaving for retirement in Western Australia. This includes a discussion of key aspectsof the project including: definition of the research question; identifying and collectingdata; analysis; and presentation of findings. The third and fourth sections provide anassessment of some advantages and disadvantages of utilising grounded theory in thecontext of a broader economic research project.

2. Selecting grounded theory as a research methodResearch within the context of critical realism requires a mode of inference over andabove the usual forms of deductive and inductive logic, which has been explained inthe following terms:

It requires a mode of inference that takes us behind the surface phenomena toits causes, or more generally from phenomena lying at one level to causesoften lying at a different deeper one. This is retroduction. (Downward &Mearman 2005)

Lawson argues in favour of retroduction as a type of reasoning that providesexplanations of the underlying causes of observed phenomenon. While retroductionappears to have some elements in common with induction, that is, moving fromobserved events to an explanatory theory of those observations, it has a significantdistinguishing feature; retroduction is a thought operation that involves moving fromknowledge of events to an explanation of their causes. In doing this, the developmentof theory or causal explanations involves moving between different ontologicaldomains (Lawson 2003:80).

Beyond identifying retroduction as the logical basis for explanation, however, Lawsonnotes that the appropriate research methods will be determined by the specific contextof each research project. Retroduction “does not necessitate that only atomisticaccounts be contemplated. Retroductive inference per se places no restriction on thesort of explanatory conception that may be uncovered” (Lawson 2003). Thus, whileLawson provides general guidelines as to the philosophical framework in which

2

research may be conducted, specific research methods must be sought that areappropriate to each project.

Grounded theory has been posited as one particular research method that can allow aretroductive approach to research that is consistent with a critical realist approach. Itis a relatively novel research approach within the discipline of economics, althoughxamples exist where it has been applied to studies of firms and behaviouralorganisation (Finch 2002), Post Keynesian research on pricing (Lee 1998; Lee &Downward 1999) and feminist research on gendered social indicators (Austen,Jefferson & Thein 2003).

Grounded theory was developed by sociologists, and it remains more commonly usedwithin sociology and other areas of business and social research than economics(Glaser 1992; Glaser & Strauss 1967). It was developed:

…as a reaction against the extreme positivism that had permeated most socialresearch…. Glaser and Strauss challenged prevalent assumptions of “grandtheory,” the notion that the purpose of social research is to uncover preexistingand universal explanations of social behaviour (Suddaby 2006: 633).

The key feature of grounded theory is that it provides a framework for generatingconceptual theory from data. In its initial stages it is an inductive approach to theorygeneration. In comparison with the more orthodox forms of hypothesis formation anddata collection used in economics, grounded theory gives a low priority to a prioritheorising or relying upon existing theoretical models. Rather it provides a method formoving from the identification of patterns of events to theorising underlying causesfor observed events and aims to produce theory with explanatory power. This aspectof the method appears largely consistent with the retroductive approach advocated byLawson.

It should be noted that the term “grounded theory” has been used to describe somecontrasting approaches to theory development. There is disagreement, even amonggrounded theory’s originators, about its correct application (Glaser 1992). Thespecific approach adopted in this project was that advocated by Glaser. While Glaserhas provided extensive detail on how grounded theory may be applied to an area ofresearch (Glaser 1978,1992,1998,2002,2003; Glaser & Strauss 1967), the key processfor moving between data collection, hypothesis formation and theoreticaldevelopment can be summarised as (Glaser & Holton 2004): Identification of an area of interest, or a broad research question, for

investigation.

The collection of data and its almost immediate analysis without the

preconception of a definitive hypothesis.

The comparison of pieces of data and the grouping together of data that has

similarities. This process is usually referred to as “coding” and results in the

formation of “categories” of data.

3

The increasing conceptualisation of data as a project proceeds to allow the

development of theoretical codes from which to develop explanations about

the relationships between categories of data.

The identification of a core variable that relates different categories and

explains variations and patterns in the data.

The use of selective coding and delimiting to focus on data and concepts

relevant to the core variable and emerging theory.

It should be noted that the key processes outlined above are iterative. Data collectionand analysis are undertaken simultaneously, with each part of analysis processinforming the direction of future data collection. However, the gradual progressionfrom open coding, to theoretical coding and selective coding reflects the shift from aninitially broad research question to a more defined research agenda that is relevant toan emerging theory. Throughout this process the research becomes more focused onspecific hypotheses and research questions; this is known as delimiting. Theresearcher is an active element in this process, making key decisions about whichcategories of data to focus on and the meaning ascribed to particular categories of data(Suddaby 2006: 638).

3. The Research ProjectIn 1992 Australia introduced a system of compulsory savings through legislationrequiring that employers pay a percentage of employees’ earnings to a specific formof savings account. In Australia this is generally referred to as a compulsoryoccupational superannuation system, although for those in overseas jurisdictions itmight be described as a private pension plan. The system is largely based on definedcontribution schemes whereby the risks associated with the returns to the fundsinvestment are largely borne by private account holders. It is expected that the fundssaved through compulsory occupational superannuation will alleviate budgetarypressures that might arise from increased government outlays on public pensions asAustralia’s demographic profile ages over coming decades.

The occupational nexus on which Australia’s superannuation scheme is based meansthat the scheme is not gender neutral in its application. Women’s broken patterns ofemployment, together with lower average wages means that they are entitled to loweremployer contributions and will accumulate lower life time levels of compulsorysuperannuation savings.

In Australia and overseas, the increasing policy priority given to private forms ofretirement saving such as occupational superannuation has lead to a range of studiesexamining relationships between gender and patterns of saving. These studies havegenerated a range of insights into the levels and sources of women’s retirementsavings and access to resources in later life (summary of Australian literature inJefferson 2005).

However, a limited number of theoretical approaches have been applied to studiesabout this subject. Typically they involved one of two broad approaches. The first has

4

been to conduct to a gender analysis of public policy in a specific jurisdiction and topredict possible scenarios for women as they grow older. The second has been toinvestigate specific hypotheses developed from a neoclassical lifecycle approach toconsumption and savings decisions. A third, emerging area of research has focusedon models of household bargaining to gain additional insights. While the approachesused so far pose particular advantages and disadvantages, their limitations mean thatsignificant gaps remain in our understanding of the causes of women’s relativelyunfavourable access to economic resources in later life. An examination of previousresearch indicates that our understanding of women’s decisions may be extended byutilising research methods that can accommodate the possibilities that: Patterns of consumption across a life-time as well as total consumption may be

important factors in consumption and savings decisions.

Monetary income is not the only relevant input of economic resources to

households.

Motivations and decision-making processes as well as wage income may be

influential in determining patterns of saving.

Savings patterns might be determined by networks of constraints rather than

the outcome of preferred choices.

Social institutions can be altered intentionally by human action to change the

constraints and choices that define economic decisions.

Significant aspects of women’s lives may be neglected or under-theorised in

existing economic models.

The goal of this particular project was to gain a fuller appreciation of the way inwhich economic, institutional and attitudinal factors operate and interact to influencewomen’s patterns of saving for retirement. In contrast with many previous studies ofwomen’s retirement savings, the project was not designed to verify the validity of anexisting hypothesis or pursue a narrowly defined research question. In addition, it washoped to identify whether there are some important issues which are not currentlyrecognised within the literature relevant to women’s retirement incomes. The researcharea in this context was therefore framed as being:

To extend our understanding of Australian women’s decisions about savingand retirement within the broad context of the experiences and institutionsthat inform and constrain those decisions.

Framing the research project in this manner had significant implications for the waydata was collected and analysed. These issues are now considered in detail.

Data CollectionGrounded theory is often associated with qualitative research and, in particular, datacollection through interviewing. Despite this association, grounded theory can utilisea range of different data, both qualitative and quantitative (Finch 2002; Glaser1998:42-3; 2003; see also Lee 2002). Within grounded theory, the collection of

5

specific forms of data is justified by reference to the suitability of data to a specificresearch question.

The issues examined in this research project were not readily accessible throughexisting quantitative data collections. The research question involved the discussionof meanings (in particular of what “saving” may mean) and the relating ofexperiences over time. Issues relating to the complexity and dynamic nature of socialstructures which may impact on women’s retirement savings decisions were alsorelevant.

Interviews and conversations which facilitate the collection of detailed qualitativedata appeared particularly suited to these questions. While these methods ofqualitative data collection are not common place within economics (Jacobsen &Newman 1997), they have been recognised as providing a method of clarifyingconcepts used in quantitative analysis and for understanding processes underlyingeconomic outcomes (Berik 1997; Pujol 1997). Further, “an obvious way to learnabout motives, constraints and the decision-making process is to ask decision makersabout them” (Bewley 2002:343). In some cases, rich qualitative data may challengeinsights gained by quantitative researchby enhancing our understanding of concepts of power, individualism and preferenceformation (Olmstead 1997). Further, interviews and conversations are consistent withfeminist, interpretive research methods as they provide a way of articulating andrecording women’s own accounts of their life experiences (Hirschfeld 1997; Oakley1995). Information about household financial arrangements has been recognised asparticularly sensitive and confidential within Australian society and in previousstudies this has resulted in a reluctance by participants to openly discuss such issues(Olsberg 1997; Singh 1997). For this reason, individual interviewing, with anemphasis on confidentiality, was adopted as the specific data collection method usedin this study.

Interview processThe interviews carried out in this project were semi-structured and relatively informal.This was done so that participants could raise issues important to them and to allowthe interview schedule to be adapted to the insights and interests of each participant.Any questions participants had about the project or personal questions of theinterviewer were answered. In keeping with previous research on women interviewingwomen, the communication process was conversational, with the two way nature ofinformation flows acknowledged in the flexible nature of the interview prompts(Hirschfeld 1997; Oakley 1995). This format meant that women who took part ininterviews shaped, to varying degrees, the content and structure of their owninterviews. For this reason, women who took part in an interview are referred to as“participants” in this research, in preference to terms which imply a more passiverole, such as interviewee or subject.

The initial interview schedule was designed following a review of previous researchabout women’s retirement incomes. It was designed to have two main sections. Thefirst section comprised an initial discussion that covered demographic details such asage, place of birth and marital status and revealed major work and householdmilestones in the participant’s life experiences such as education, workforce entry andexit, occupation, household formation, child rearing, marriage, divorce and future

6

expectations in these areas. This discussion was initiated with the question, “Couldyou give me a five or ten minute version of your life story?” The aim of thediscussion was for participants to identify the experiences they perceived as mostsignificant in their life course. In gathering this particular data, the aim was to gaininsights into the interactions between different life course events and experiencesregarding labour supply, earning an income and savings. Specific questions from theresearcher within this section of the interview were confined to matters of clarificationof the participant’s story, for example, querying the participant’s age at the time ofparticular events. The opening question generally put participants at ease with theinterview process. In particular, this style of question appeared to make it clear thatthere were no “right” or “wrong” answers to the question and participants had a highdegree of control over the content of their answers. The research topic was also onethat participants appeared to feel knowledgeable about and confident answering.

The second section of the discussion focused more specifically on approaches tosaving and retirement planning. Again, the goal was to allow for the emergence ofissues perceived by the participant as significant and for participants to define theactions they believe constitute savings and retirement planning. Some examples ofinterview prompts are: Can you tell me about how you manage money in your household?

Can you tell me about your strategies to accumulate savings?

Can you describe how you keep your savings?

Can you tell me about things which have made it very difficult or easy to

save?

What thoughts have you had about saving for retirement?

How do you think you will meet your financial needs in old age?

The above schedule was piloted in a set of three interviews. Initial analysis revealedthat the schedule facilitated a relatively free flowing discussion that generatedsignificant, although complex, data relevant to women’s life experiences and theirapproaches to savings and retirement. It was also found that it was sometimesunnecessary to specifically ask every question, as participants often covered particularissues without prompting. The main finding from the pilot program of interviews wasthe desirability of ensuring that participants had sufficient time to consider anddiscuss particular prompts before further questions were asked. It was decided tomaintain the initial interview schedule for the remainder of the study.

Interview locationInterviews were held at a location and time which was convenient and comfortable forthe participant. Some participants had a strong preference for meeting somewhere‘neutral’, while others preferred to meet at their own home or to meet at the universitycampus. This meant that interviews were conducted at private homes, workplaces,coffee shops, a local park or offices at the university.

Payment of participants

7

At the conclusion of the interview, participants were paid $40 to compensate them forexpenses and inconvenience incurred in attending an interview. This money was madeavailable through funding provided by two state government offices: The Office forWomen’s Policy and the Office for Senior’s Interests and Volunteering, together withfunding from the Australian Research Council through a Linkage Project(LP0347060).

Participant recruitment and selectionUsing a grounded research approach means that data should be collected within aframework of theoretical sampling rather than through the process of statisticalsampling more commonly used in economics. Statistical sampling is used to test orverify existing theory and to make inferences about the extent to which therelationships between identified variables hold in a specified population. In contrast,theoretical sampling is used to establish an iterative process between theorydevelopment and data collection. At the outset, identified variables and causalrelationships are not defined in grounded theory research. Initial sampling is carriedout with the aim of developing themes and categories directly from data. Furthersampling is then carried out to facilitate a knowledge of the applicability of thesethemes and categories to further cases (Finch 2002).

Recruitment was designed to facilitate the collection of data from a heterogenoussample of participants. This process of selection facilitates the collection of two typesof data: high quality case descriptions, which document uniqueness; and commonexperiences across participants (Morse 1994). Selecting this form of sampling wasconsistent with the researchers’ interest in studying areas of commonality anddiversity within a varied population.

In the context of Western Australia, which is a large, culturally and geographicallydiverse state, some initial “areas of diversity” were identified which could possiblyimpact on women’s motivation and ability to save for retirement. Five areas ofdiversity were identified: socio-economic background; cultural background; age orstage in the life cycle; geographic location; and attachment to the labour market.However, these identified areas of diversity are not exhaustive and do not include, forexample, sexuality and household structure. It was recognised that this meant the datacollection process may require modification if it became apparent that insights couldbe gained from participants with characteristics not identified at the commencementof the research program. In keeping with grounded theory methods, data collectionand analysis necessitated flexibility with respect to decisions about the collection ofdata throughout the project.

Following initial analysis of three pilot interviews, approximately 150 invitations toparticipate in the research project were distributed in two rounds of approximately 75invitations each. Invitations were distributed to the parents of children in three classesat two different primary schools attended by children aged approximately between sixand 12 years; to parents of children aged between one and four years attending a pre-school playgroup; to members of two community-based multicultural social groups; asocial walking/exercise group of women who meet on a weekly basis; and an officewhich employed a number of women in full time, part time and causal positions. Afurther two invitations were issued in response to requests from women whocontacted the university because they had heard about the research project through

8

friends or colleagues and wished to participate. The groups were chosen to provide anopportunity for women from diverse socioeconomic backgrounds to participate in theproject. The invitation broadly outlined the subject and goals of the research andprovided a brief description of the interview process.

An analysis of selected socio-demographic characteristics confirmed diversity interms of age, occupation, workforce participation, household structure, number ofchildren, marital status and country of birth. Despite this diversity, it is possible toidentify some populations that are unrepresented in this sample. For example, therewere no participants who identified themselves as being aboriginal and all participantshad stable places of residence. This means that it is likely that there are somesignificant perspectives on saving and retirement that are not represented in this study.

When using grounded theory, adequate theoretical sampling is assessed in terms ofwhether further data collection no longer yields results which are surprising orproviding exceptions to previously developed themes and categories. This is termed“saturation” and implies that there is a limited scope for further understanding toemerge from continued sampling. As noted above, constraints of time and fundingmeant that the project was completed before saturation was achieved. Asdemonstrated in the following chapter, some categories contained a great deal ofrecurring information about experiences and perceptions. In contrast, there werecategories that contained relatively small amounts of data. This is discussedthroughout the following two chapters.

Ethical considerationsThe most significant ethical issue for this project was the need to ensure participantconfidentiality. Previous Australian studies of household financial arrangements havedemonstrated that participants find the discussion of household money and financialdecisions to be a particularly sensitive issue (Edwards 1984; Singh 1997). The abilityto collect high quality data in an ethical manner was therefore closely aligned with theneed to ensure that data could not be matched with identified participants. However,records of participant names and contact details were required for two reasons:Firstly, to facilitate arrangement of the interview and secondly, to arrange delivery ofan interview transcript to the participant for verification and, if necessary, alteration.To facilitate these arrangements, transcripts included a non identifiable code whichcould be linked to a list held only by the researcher and the project supervisors. Thislist was a hard copy kept in a locked office cabinet. In addition, any identifyinginformation within the transcript was altered in such a way as to protect the identity ofthe participant while preserving the meaning of the discussion.

The written invitations informed participants of the objectives of the project and ofthe intention to record discussions for transcription and analysis. These details wereconfirmed in writing when interviews were arranged and the written consent of eachparticipant was sought before any interviewing or recording proceeded.

Following analysis of interview data, a summary report outlining the main issuesdiscussed in the interviews was sent to each participant, together with a letter thankingparticipants for their input in June 2005. All relevant details of the data collectionprocess were described in an application to Curtin University of Technology’s Human

9

Research Ethics Committee. Approval to proceed with the project was granted priorto the first pilot interview proceeding and continued for the duration of the project.

Data formatIt should be noted that Glaser cautions against taping and transcribing interviews aspart of the data collection and analysis process and expresses a preference for relyingon field notes (Glaser 1998). In this project however, a conscious decision was madeto tape and transcribe interviews for two reasons. Firstly, it was felt that taping andtranscribing interviews would increase confidence in the data collection processamong those assessing funding applications. Secondly, it was decided that thisprocess might increase the confidence of participants in the research. In particular, itwas felt that participants should be given a record of the interview and an opportunityto alter and clarify the views that it contained.

The decision to tape and transcribe interviews meant that transcriptions formed themain body of the data analysed in this project. However, after each interview, theresearcher completed field notes that, in addition to details about the time and place ofthe interview, canvassed thoughts about the issues discussed throughout eachinterview. These memos and others written throughout the project assisted the dataanalysis process, as described below.

Data Management and AnalysisThe interview transcripts and field notes yielded approximately 650 pages of text.This comprised the data set for the project and was managed using N*Vivo softwarewhich has been developed specifically to facilitate qualitative data analysis.

Data analysis commenced following the first interview. Upon completion of the firsttranscript, open coding was commenced by reading through the transcript and notingthe various issues and experiences discussed by the participant. It was intended thatthis process would be repeated for subsequent interviews, with each being transcribedand subject to an analysis before the next interview proceeded. The aim of thisprocess was to implement an iterative approach to data collection and analysis.However, at about the mid-point of the interview schedule, it became impractical tocomplete transcripts prior to the next interview taking place. This occurred because ofthe constraints of accommodating the wishes of participants with respect to the timingof their interview. Most participants wished for, or expected, their interview to be heldreasonably soon after posting their reply. At various times during the project, therewere up to six interviews to be transcribed when subsequent interviews were held.

A conscious decision was made to allow taped interviews to accumulate in preferenceto spacing interviews to accommodate the requirements of the researcher. Many of theparticipants’ lives were busy and it became apparent that the capacity and willingnessof people to participate in the project could change quite quickly. For example, oneparticipant experienced an accident that required her to undergo extensive medicaltreatment and this resulted in her withdrawal from the project. Some other participantswere pregnant and it suited them to have their interviews scheduled prior to the birthof their baby. Other significant examples of changes in participants’ circumstancesincluded the death of a parent (in the case of two participants), moving house and animpending marriage and holiday.

10

Within these constraints, analysis of each interview transcript continued alongside thedata collection process. Analysis commenced with the generation of categories fromthe transcript data. This was done through a process of open coding. Open codingproceeds by developing categories of different data, for example by grouping togetherall participants’ comments about allocating to children’s savings accounts, but it doesnot assign priority to specific categories, nor define relationships between them (Dey1999; Finch 2002; Glaser 1992; Miles & Huberman 1994).

In developing categories of data, priority was given to Glaser’s guidelines that data“be closely examined and compared for similarities and differences, while constantlyasking of the data the neutral question “What category or property of a category doesthis incident indicate?” (Glaser 1992:39, emphasis in original). Throughout thisprocess it was recognised that categorisation involves an interpretation andconceptualisation of the data by the researcher (Dey 1999:252-260; Finch 2002:202).However, the intention throughout the initial analysis was to focus on the content ofparticipants’ comments and to develop a system of categories that reflected theirexperiences and perceptions relevant to savings and retirement. That is, to develop“data driven” categories (see also Lee 2002:794-5; Richards & Richards 1995:80).

While the development of categories requires interpretation and conceptualisation bythe researcher (Dey 1999), interpretation of data became even more apparent whencategories that dealt with similar issues were grouped together to assist analysis.There are several different terminologies that describe the grouping of categories ofdata. For example, Glaser uses the terms “concepts”, “properties” and “categories”(Glaser 1992:38). Within the context of N*Vivo software, different levels of groupingare referred to as “nodes” or “trees”. Other terminology exists, such as “children”,“siblings”, “trees” and “roots” (Richards & Richards 1995). For the purposes ofanalysis in this project, groupings of categories were labelled “constructs”, reflectingthe active role of the researcher in identifying specific categories as related and thenconstructing labels, to group them together. This part of the process reflected thetheoretical coding described by Glaser. During this part of analysis there wasrelatively less use of computer aided data management. Instead, memos and fieldnotes were utilised and more memos generated, as possible constructs wereconsidered. The development of constructs continued simultaneously with the processof writing drafts of the research findings, as the writing process also provided anopportunity to consider and assess different possible constructs.

Finally, different constructs were grouped together into broad concepts. Thisrepresented another level of abstraction and interpretation. Three broad concepts weredeveloped in order to develop a theoretical discussion from the collected data. Adetailed description of the concepts, constructs and categories generated from the dataand used as the basis for organisation and theorising is given in the following chapter.

Criteria for assessing causal explanationsA final part of the research design for this project was the need to specify criteria thatcould be used to assess the adequacy of the data collection and analysis process. Thisimportant issue poses some challenges because appropriate criteria for assessing theoutcomes from grounded theory research method are rarely discussed in economicliterature. As grounded theory is not a process of theory verification and uses aprocess of theoretical sampling, the adequacy of data collection and analysis cannot

11

be based on “traditional” assessments such as the representative aspects of the data orrelationships between data and hypotheses. For the purposes of this project, criteriawere developed using arguments from Downward, Finch and Ramsay (1999) andRunde (1998).

An overview of the project findingsWhen the data had been categorised and conceptualised into a relatively integratedframework, the stories that emerged about women’s savings were considerably moredetailed than the stories typically told through orthodox economic modelling. This, ofcourse, was not particularly surprising, given the contrast between the relevantresearch methods. As explained above, part of the rationale for this project was thewish to investigate the types of issues that might be omitted from current analyses. Akey area of interest, therefore, was to compare and contrast the findings from theproject with existing models of savings decisions.

It was clear that some of the transcript data was relevant to economics’ frequentlymodelled intertemporal choice problem: the difficulty of choosing betweenconsumption now and consumption later (Deaton 1992). In addition, there werecategories of data that appeared relevant to hyperbolic discounting, dynamicpreference modelling (see for example Angeletos, Laibson, Repetto, Tobacmann &Weinberg 2001; Laibson 1997,1998) and procrastination and retirement planningdecisions (O'Donoghue & Rabin 1999a,b,2001). However, while these approachesappear to have some overlap with the data collected in this study, their focus onindividual decisions, transactions costs and changing preferences, meant that theirexplanations of saving behaviour omitted some significant aspects of the findingsfrom in this study.

One of the main contrasts between previous accounts of retirement savings decisionslay in the relative emphasis to given to individual decisions based on expectedoutcomes. The data collected in this project suggest that the social contexts in whichwomen make decisions about retirement savings have important effects on thedecision-making processes that are used and ultimately, on the types of decisionsmade. In comparison, there was relatively little emphasis given to specific, plannedoutcomes for accessing income in later life.

Participants described a number of financial routines that had become establishedwithin their households and assisted with the ongoing need to allocate income tospecific purposes. The routines ranged from processes which allowed little role forsaving, to those which allowed for saving from “residual” income, to those thatprioritised saving through the use of special accounts. Throughout the descriptions ofallocating income to expenditure or savings however, participants described their useof routines which prevented the need for overt, regular decision-making.

One of the striking features of many transcripts was the lack of a link between currentsavings actions and identifiable, long term plans for accessing income in retirement.There was little in the data to suggest that eventual retirement income was estimated,that different investment vehicles were considered in detail or that varying outcomesfor retirement were considered. Rather, most data appeared relevant to theestablishment of routines that negated the need for regular, active decision-making. Inthis context, buying housing was seen as a particularly desirable strategy: regular

12

payments are required, everyone else does it and it is familiar – most people knowwhat a house is and what some of its benefits are. Housing was also viewed as aparticularly safe investment, with one participant saying: “they do say as safe ashouses.

In short, the study found that women’s relatively low independent retirement savingsare the result of two levels of contributing causal factors. At an ‘intermediate’ level itcan be argued that women’s relatively low levels of independent retirement savingsare caused by both their limited access to independent incomes and householddecision-making processes that reproduce traditional savings patterns. These findingsare largely in keeping with the insights generated from previous studies of women’ssavings and retirement incomes, as outlined above.

However, the depth and richness of qualitative data meant that previously identified“intermediate” causes of women’s patterns of savings were linked with a number ofother factors, including, including: The complexity associated with long-term planning, joint decision-making and

retirement savings options;

Gender norms of workforce participation and household resource allocation;

A context of emotions that frame and limit options that are included within

household decision-making processes;

Bounded rationality; and

Interactions between individuals, particularly household members and the

effects this has on preference.

The main area of contrast between this study and the insights from previous studieswas the range of issues that were relevant to household decision-making processes,particularly those issues that extended beyond those of gender differences in marketwages and life expectancies. Further, it was found that the current institutional contextfacilitates the continued implementation of traditional decision-making routines thatare likely to contribute to continued suboptimal outcomes associated with under-saving.

It was theorised that the causal mechanism underlying women’s low, and potentiallysuboptimal, retirement savings is a process consisting of two steps. Firstly, specificfeatures of women’s decision-making contexts facilitate the development of decision-making routines or compromises. Secondly, those routines involve the developmentof processes which give inadequate attention to their likely outcomes, particularlythose outcomes which affect women’s capacity to access economic resources inretirement. This does not mean that all savings decisions will be inadequate to meetindividual retirement needs. Particular individuals or households may have sufficientforesight and skills to address the challenges of the current decision-making context.The argument is that the current context of retirement planning systematicallyaccommodates the implementation of suboptimal decisions, not that it necessitatesthem.

13

The key features of the decision-making framework conceptualised in this chapter arelisted below in Table 1. The first column lists labels developed to describe keyfeatures of women’s financial decision-making context. The second column providessome details of the identified features of that context. The third column lists theimplications of these features and summarises why implications may apply differentlyto women as a population group, compared with men. That is, in addition tohypothesising that the existing retirement incomes framework facilitates under-saving, it is also argued that it is not gender neutral in its implications and thisprovides some insights into the different savings patterns observed between men andwomen. Particular features of the decision-making context that accommodate under-saving for retirement are overcome through mandatory schemes such as occupationalsuperannuation. Women are disproportionately represented in the population that fallsoutside of Australia’s current compulsory savings scheme.

The implication of these causal relationships is that women’s lower retirementincomes result from multiple causes rather than a simple relationship between lowearnings and savings. For example, the complexity of forming expectations about life-time income is greater for women because of greater heterogeneity in patterns ofworkforce participations. Similarly, gender norms of care have implications forhousehold’s financial priorities and organisation.

It should be recognised that heterogeneity in the population of men and women meansthat the contextual features summarised in this table may apply differently to specificindividuals. For example, a single woman with relatively high earnings, familiaritywith investment decisions and a relatively high focus on retirement outcomes mightbe expected to have adequate access to savings to finance retirement. Similarly, amarried man with broken patterns of employment and relatively little knowledge ofsavings and investment options may be expected to experience poor outcomes inretirement. The gender implications of the savings context do not depend on intrinsicdifferences between men and women. Rather they result from structural differences,as a group, in men’s and women’s patterns of workforce participation, access toinformation and adherence to gender norms.Table 1: Suboptimal savings contexts and some gender implications

Contextualfeature

Details of this feature Possible links to gender implications ofsavings framework

Complexity Unpredictable earnings andaccess to household income

Women earn lower earnings and haverelatively heterogenous earnings patterns.This may be exacerbated by incompletepooling of household resources

Complex investment framework Equally applicable to men and women butcircumvented when there is an entitlement tosuperannuation. This applies more commonlyto men.

Joint decision-making processes As above. Further, women’s influence ondecision-making may be further reducedthrough low individual income.

Emotional aspects Fear of future events Equally applicable to men and women.However, women’s longer life expectanciesmean that they are more likely to face afuture on their own.

14

Decisions framed to prioritiseneeds of others, especiallychildren

Equally applicable to men and women.However, this is circumvented by access tocompulsory superannuation. Gender normsappear to operate in combination withemotions to give low priority to women’sindependent savings.

Gender and socialnorms

Limits on workforce participation Highly relevant to women who undertake alarger share of unpaid work.

Low priority on independentsaving

May be equally applicable to men andwomen. However, “breadwinner” approachesto household saving may prioritise men’ssavings. Compulsory superannuation alsocircumvents this issue.

Relevance of peer groups onfinancial decisions.

Equally applicable to men and women.

Boundedrationality

Financial skills notcommensurate with demands ofdecision-making context

Equally applicable to men and women. Ifskills are enhanced through workplaceinformation/training then this may applyespecially to women.

Linkedindividuals

Difficulty in isolating individualneeds from those of others

Equally applicable to men and women.

The development of the theoretical framework outlined in Table1, whichaccommodates the possibility that the operation of constraints may differ betweenindividuals, means that while specific constraints may facilitate the implementation ofsuboptimal saving strategies, they do not necessitate it. It remains possible forindividuals to save adequately within this framework, although this may be a morechallenging process than would be the case in alternative frameworks. Finally, theframework in Table 1 is relatively amenable to alteration to reflect insights fromadditional data or future research. It is intended as one way of identifying causes ofunder-saving and the different implications they have for men and women, however,it is unlikely to be definitive.

The conclusion of the study was that the combination of recent changes in Australia’sretirement income framework based on private savings decisions by households givesrise to a situation where women’s retirement incomes appear likely to remaininadequate and suboptimal. The term “suboptimal” in this context is used to describesavings that are below the level that a decision-maker would prefer if they were ableto form reasonable expectations about the likely outcomes of their savings decisions.It does not refer merely to the fact that there may be a reduction in consumption afterretirement because this may have been foreseen and accepted.

4. Some disadvantages of using grounded theoryThe focus of grounded theory on the generation of new theory has been identified asproblematic in some circumstances. For practical reasons, particularly thoseconcerned with time and funding, it is not always be possible to continue collectingfurther data until “saturation” of a specific category occurs (Glaser & Strauss 1967;Morse 1994). Saturation is a term that is linked with grounded theory’s use oftheoretical rather than statistical sampling. In brief, it refers to a point in the datacollection process where no new or surprising categories of data are emerging. This,

15

in turn, means that the theoretical explanations generated through the theorydevelopment process apply to a wide range of relevant data. Without saturation, it isdifficult to confidently put forward a theory that will adequately explain all relevantaspects of a research question and generate generaliseable research findings.

Limits on specific projects, such as limited time and funding, mean that the outcomesof a particular study may result in contributions that fall short of the development of anew theory. This has lead to an examination of the potential uses of grounded theoryin contexts where such limitations are likely to exist. Finch suggests that an iterativeprocess of data collection and hypothesis formation allows the emergence “novelcontributions to knowledge” which can extend existing theory and be incorporatedinto future research projects (Finch 2002). This approach is consistent with the viewthat theory construction is part of an ongoing process rather than an activity whichallows formation of a finished product (Glaser & Strauss 1967). That is, while thegoal of research projects using grounded theory is to construct new theory, it isunlikely that a project with specified time and resources constraints will generate anall encompassing theory of savings (or any other social phenomenon). Suchconstraints apply to the context of this research. However, within these constraints, itmay be possible to produce insights or “novel knowledge claims” (Finch 2002) whichextend or challenge current theoretical constructs.

It is unlikely that saturation was reached in this project. There were a range ofcategories with relatively few data and it appeared likely that ongoing data collectioncould generate additional insights. Prior commitments to funding bodies andbudgetary constraints made the data collection process relatively constrained. Theoriginally projected twenty interviews were increased to thirty during the life of theproject and this facilitated the collection of further data. However, there was littlecapacity to change direction and for example, implement a survey based on thequalitative research findings. These are not necessarily difficulties with groundedtheory but they are challenges within research project guidelines that are based on theassumption that research will test specific hypothesis and have predetermined datacollection goals and guidelines.

A further, and related point, relates to the satisfaction of funding research partnerswith the outcomes of a project such as this. In the case of this particular project, thepersonnel involved in the original funding application were closely involved with thedesign of the project and had a reasonably clear understanding of the project’s likelyoutputs. However, there were changes in personnel at the relevant State Governmentoffices over the life of the project. At the conclusion of the project there was somedegree by at least one official about the research methods were used (nonrepresentative sample for example). Such challenges appear closely related to theextent to which a researcher’s interpretation of data clearly shapes data interpretationand analysis. For those familiar with grounded theory this isn’t seen as a particularlyproblem:

Successful grounded theory has a clear creative component. Glaser andStrauss were aware of this component and the tension it would create withthose who find comfort in trusting an algorithm to produce results…. Thosenew to grounded theory research must become both patient and tolerant ofambiguity, because it is the ongoing interaction between researcher and data

16

that generates the fundament of successful grounded research (Suddaby 2006:638).

While these issues can be addressed, they do illustrate that the relatively lack offamiliarity among research partners can lead to concerns that may not arise whenmore conventional or familiar research methods are employed.

A similar point can be made with respect to discussions about the findings of thisproject with colleagues in economics departments. A lack of familiarity withmethodological debates, particularly discussions taking place within heterodoxschools of thought can lead to varying degrees of confusion among colleagues as towhy this type of research should be undertaken. At the same time, however, it doesprovide an opportunity for discussing the design of research design. This can be bothchallenging and rewarding. However, the experience of this project suggests thatsome of the more rewarding comments and questions may arrive several days orweeks after the initial presentation of findings.

5. Some advantages of adopting a grounded theory approachIn the context of this project, however, grounded theory also offered severaladvantages. Firstly, it provided some indication that some significant aspects ofwomen’s experiences with respect to savings strategies remain under-theorised andthat there are difficulties with only applying existing theoretical constructs in amanner that can potentially extend our understanding of processes and events relevantto women’s savings decisions. As a process of theory development, grounded theoryoffered one way of generating new insights into this subject. The process allowed fornovel categories of data to emerge during the data collection process and forpotentially new causal processes to be theorised.

The findings from this project were interesting for at least two reasons. Firstly, asoutlined above, they indicated that there are some factors unrelated directly towomen’s relatively lower levels of earnings and relatively higher life expectanciescould be systematically relevant to their levels of retirement savings. Secondly,however, the study provided a mechanism for the identification of a range of existingeconomic literature that could potentially provide further insights into women’spatterns of saving and relatively low access to resources in later life.

For example, while these findings were developed directly from the data, they wereconsistent with existing literature, identified following the data collection and analysisprocesses, that link specific decision-making processes, particularly habits and rules,with contextual features such as uncertainty, complexity, extensiveness and emotions.The importance of these links is that, while not previously applied to studies ofwomen’s savings, there is a large literature suggesting that a complex institutional orregulatory context encourages the use of decision-making “short cuts” or “habits” thatmay not necessarily allow for optimal outcomes (Davidson 1987; Hodgson 1997).The development “mental accounts” and the non-fungibility of household resourcesmight be particularly relevant in this context and while these concept has been appliedto various areas of financial decision-making they have rarely been used with respectto women’s savings decisions (some interesting examples of applying these conceptsto economic research are provided by McGraw, Tetlock & Kristel 2003; Winnett &Lewis 1995). Further, in contexts where emotions are important, decisions will be

17

framed in such a way that particular options may not be considered (Elster 1996,1998;Nelson 2003). This might be particularly relevant to the context of householddecision-making although it as an area of economic theory remains largelyundeveloped.

Perhaps not surprisingly, the findings were also consistent with a range ofinternational studies that identify gender norms as significant in the management ofhousehold resources (Edwards 1984; Pahl 1989,1995; Singh 1997; Vogler 1998;Vogler & Pahl 1994). Similarly, there are diverse approaches to showing the difficultyof separating financial decisions from the relationships in which they take place(Ingham 1999; Nelson 1994; Zelizer 1994a,b,2000). This suggests that whiledifferences in income are one important cause of differences in women’s and men’slevels of savings, it is also possible that different gender expectations about thefinancial decisions and roles played by men and women could play an important role.This might be an avenue for investigation that sheds further light on apparentdifferences in risk aversion that have been noted in some literature from the UnitedStates, as discussed above.

The above list of potentially relevant studies and theoretical approaches not yetapplied to women’s approaches to saving for retirement could be extended. However,the main point is that a relatively small qualitative study can provide the impetus foridentifying potentially relevant empirical and theoretical approaches that already existwithin the large body of economic literature that is available to researchers. Suchapproaches can be compared against the small amount of data collected in thequalitative approach and assessed for their potential application and explanatorypower to specific questions that have been identified as particularly relevant.

In summary, adopting a research method, such as grounded theory, that contrasts withprevious studies that address a particular economic research question can have at leasttwo types of benefit. The first relates to the wide range of data and explanation thatcan be used to gain insights into a particular research question. A second benefit is thecapacity for “new” findings to be the source of developing links with theoretical andapplied investigations in other subject areas of economics. In a discipline as variedand large as economics this can provide a relatively efficient way of identifyingpotentially relevant research methods and insights that might not have appearedimmediately relevant to a specific research topic.

6. ConclusionTwo key conclusions might be drawn from this example of using grounded theory toaddress an economic research project. Firstly, there are considerable challenges to bemet when employing grounded theory in an economic research project. Thesechallenges arise from both the history and nature of grounded theory itself and fromthe relative lack of understanding among many economists of the goals and processesinvolved in grounded theory research projects. These challenges are notinsurmountable but they can be considerable and often go the question of the extent towhich specific institutional practices such as funding guidelines, presuppose acommitment to particular approaches to research. Secondly, and perhaps moreimportantly, grounded theory can provide a useful way of generating potentiallyuseful insights that are not generally obtained from the more standard array ofeconomic research methods. As such it might be viewed as one method that can

18

contribute to the plurality of approaches that can be utilised to gain insights in to thecausal mechanisms that generate our observations about our economic environment.

ReferencesAngeletos, G.-M., Laibson, D., Repetto, A., Tobacmann, J. & Weinberg, S. 2001,

'The hyperbolic consumption model: Calibration, simulation and empiricalevaulation', Journal of Economic Perspectives, vol. 15, no. 3, pp. 47-68.

Austen, S., Jefferson, T. & Thein, V. 2003, 'Gendered social indicators and groundedtheory', Feminist Economics, vol. 9, no. 1, pp. 1-18.

Berik, G. 1997, 'The need for crossing the method boundaries in economics research',Feminist Economics, vol. 2, no. 3, pp. 121-25.

Bewley, T. 2002, 'Interviews as a valid empirical tool in economics', Journal ofSocio-Economics, vol. 31, no. 4, pp. 343-353.

Davidson, P. 1987, 'Sensible expectations and the long-run non-neutrality of money',Journal of Post Keynesian Economics, vol. 10, no. 1, pp. 146-153.

Deaton, A. 1992, Understanding Consumption, Clarendon Press, Oxford.

Dey, I. 1999, Grounding Grounded Theory: Guidelines for Qualitative Inquiry,Academic Press, San Diego.

Downward, P. & Mearman, A. 2005, 'Retroduction as mixed-methods triangulation ineconomic research: Reorienting economics into social science'.

Edwards, M. 1984, The income tax unit in the Australian tax and social securitysystems, Australian Institute of Family Studies, Melbourne.

Elster, J. 1996, 'Rationality and the Emotions', The Economic Journal, vol. 106, no.438, pp. 1386 - 1397.

Elster, J. 1998, 'Emotions and economic theory', Journal of Economic Literature, vol.36, no. 1, pp. 47-74.

Finch, J. 2002, 'The role of grounded theory in developing economic theory', Journalof Economic Methodology, vol. 9, no. 2, pp. 213-234.

Glaser, B. 1978, Theoretical Sensitivity, Sociology Press, Mill Valley, California.

Glaser, B. 1992, Basics of Grounded Theory Analysis, Sociology Press, Mill Valley,California.

Glaser, B. 1998, Doing Grounded Theory: Issues and Discussion, Sociology Press,Mill Valley, California.

19

Glaser, B. 2002, 'Conceptualization: On theory and theorizing using grounded theory',International Journal of Qualitative Methods, vol. 1, no. 2, p. Article 3.

Glaser, B. 2003, The Grounded Theory Perspective II: Description's Remodeling ofGrounded Theory Methodology, Sociology Press, Mill Valley CA.

Glaser, B. & Holton, J. 2004, 'Remodeling grounded theory', Qualitative SocialResearch, vol. 5, no. 2, p. article 4. Retrieved: 24 June 2005, fromhttp://www.qualitative-research.net/fqs-texte/2-04/2-04glaser-e.htm database.

Glaser, B. & Strauss, A. 1967, The Discovery of Grounded Theory, Aldine, Chicago.

Hirschfeld, M. L. 1997, 'Methodological stance and consumption theory: A lesson infeminist methodology', in JB Davis (ed.), New Economics and Its History,Annual Supplement to Volume 29, History of Political Economy, DukeUniversity Press, Durham, pp. 191-211.

Hodgson, G. M. 1997, 'The ubiquity of habits and rules', in GM Hodgson (ed.),Recent Developments in Institutional Economics, Edward Elgar, Cheltenham,UK, pp. 379-400.

Ingham, G. 1999, 'Money is a social relation', in S Fleetwood (ed.), Critical Realismin Economics, Routledge, London, pp. 103-1224.

Jacobsen, J. P. & Newman, A. E. 1997, 'What data do economists use? The case oflabor economics and industrial relations', Feminist Economics, vol. 3, no. 2,pp. 127-130.

Jefferson, T. 2005, 'Women and retirement incomes in Australia: A Review',Economic Record, vol. 81, no. 254, pp. 273-291.

Laibson, D. 1997, 'Golden eggs and hyperbolic discounting', Quarterly Journal ofEconomics, vol. 112, no. 2, pp. 443-477.

Laibson, D. 1998, 'Life cycle consumption and hyperbolic discount functions',European Economic Review, vol. 42, pp. 861-871.

Lawson, T. 2003, Reorienting Economics, Routledge, London.

Lee, F. S. 1998, Post Keynesian Price Theory, Cambridge University Press,Cambridge.

Lee, F. S. 2002, 'Theory creation and the methodological foundation of PostKeynesian economics', Cambridge Journal of Economics, vol. 26, no. 6, pp.789-804.

Lee, F. S. & Downward, P. 1999, 'Re-testing Gardiner's Means's evidence onadministered prices', Journal of Economic Issues, vol. 33, pp. 861-886.

20

McGraw, A. P., Tetlock, P. E. & Kristel, O. V. 2003, 'The limits of fungibility:rational schemata and the value of things', Journal of Consumer Research, vol.30, no. 2, pp. 219 - 229.

Miles, M. B. & Huberman, A. M. 1994, Qualitative Data Anaysis: An ExpandedSourcebook, Second Edition, Sage, Thousand Oaks.

Morse, J. M. 1994, 'Designing funded qualitative research', in NK Denzin & YSLincoln (eds.), Handbook of Qualitative Research, Sage, Thousand Oaks,California, pp. 220-235.

Nelson, J. A. 1994, 'I, thou and them: Capabilities, altruism, and norms in theeconomics of marriage', American Economic Review Papers and Proceedings,vol. 84, no. 2, pp. 126-131.

Nelson, J. A. 2003, 'Once more with feeling: Feminist economics and the ontologicalquestion', Feminist Economics, vol. 9, no. 1, pp. 109-118.

Oakley, A. 1995, 'Interviewing women: A contradiction in terms', in H Roberts (ed.),Doing Feminist Research, Routledge, London and New York, pp. 30-61.

O'Donoghue, T. & Rabin, M. 1999a, 'Doing it now or later', American EconomicReview, vol. 89, no. 1, pp. 103-124.

O'Donoghue, T. & Rabin, M. 1999b, 'Procrastination in preparing for retirement', inHJ Aaron (ed.), Behavioral Dimensions of Retirement Economics, BrookingsInsitution, Washington DC.

O'Donoghue, T. & Rabin, M. 2001, 'Choice and procrastination', The QuarterlyJournal of Economics, vol. 116, no. 1, pp. 121-161.

Olmstead, J. C. 1997, 'Telling Palestinian women's economic stories', FeministEconomics, vol. 3, no. 2, pp. 141-151.

Olsberg, D. 1997, Ageing and Money: Australia's Retirement Revolution, Allen andUnwin, St Leonards, NSW.

Pahl, J. 1989, Money and Marriage, MacMillan, London.

Pahl, J. 1995, 'His money, her money: Recent research on financial organisation inmarriage', Journal of Economic Psychology, vol. 16, no. 3, pp. 361-376.

Pujol, M. 1997, 'Broadening economic data and methods', Feminist Economics, vol. 3,no. 2, pp. 119-120. Retrieved: 27 June 2005, from SwetsWise database.

Richards, T. & Richards, L. 1995, 'Using Hierarchical Categories in Qualitative DataAnalysis', in U Kelle (ed.), Computer-aided Qualitative Data Analysis:Theory, Methods and Practice, Sage Publications, London.

21

Runde, J. 1998, 'Assessing causal economic explanations', Oxford Economic Papers,vol. 50, no. 2, pp. 151-172.

Singh, S. 1997, Marriage Money, Allen and Unwin, Sydney.

Suddaby, R. 2006. ‘From the Editors: What grounded theory is not’ Academy ofManagement Journal vol 4, no. 4, 633-642.

Vogler, C. 1998, 'Money in the household: Some underlying issues of power', TheSociological Review, vol. 46, pp. 687-713.

Vogler, C. & Pahl, J. 1994, 'Money, power and inequality within marriage', TheSociological Review, vol. 42, no. 2, pp. 263-288.

Winnett, A. & Lewis, A. 1995, 'Household accounts, mental accounts and savingsbehaviour: Some old economics rediscovered?' Journal of EconomicPsychology, vol. 16, pp. 431-448.

Zelizer, V. A. 1994a, 'The creation of domestic currencies', Amercian EconomicReview Papers and Proceedings, vol. 84, no. 2, pp. 138-142.

Zelizer, V. A. 1994b, The Social Meaning of Money, Basic Books, New York.

Zelizer, V. A. 2000, 'Fine tuning the Zelizer view', Economy and Society, vol. 29, no.3, pp. 383-389.

22